Abstract
The surge in artificial-intelligence (AI) infrastructure investment raises a financial question distinct from the productivity effects of AI: can the firms financing the buildout capture enough of the value created by AI as cash flow to recover their investment? This paper develops a discounted-cash-flow framework that separates economy-wide value creation from investment recovery by those firms. The framework is applied to the reported and projected 2024–2027 capital expenditures of Amazon, Alphabet, Microsoft, Meta, and Oracle. Under the baseline 20-year U.S. calibration, recovering the full reported 2024–2027 capital-expenditure benchmark requires the long-run level of GDP to be approximately 19 percent above its no-AI path under the baseline value-capture calibration. The required economic payoff rises when the capital-bearing value-capture rate is lower, diffusion is delayed, the required return is higher, the recovery horizon is shorter, or additional post-2027 expansion investment is included. It falls when a smaller fraction of reported capital expenditure enters the valuation, a larger economic base is used, or monetizable gains outside measured GDP are included. The results show that even a large economy-wide payoff does not ensure investment recovery: enough of the value created by AI must reach the firms financing the infrastructure as cash flow, and it must do so sufficiently quickly.